Meiling Li

Guobiao Li, Sheng Li, Meiling Li et al.

Steganography is a technique for covert communication between two parties. With the rapid development of deep neural networks (DNN), more and more steganographic networks are proposed recently, which are shown to be promising to achieve good performance. Unlike the traditional handcrafted steganographic tools, a steganographic network is relatively large in size. It raises concerns on how to covertly transmit the steganographic network in public channels, which is a crucial stage in the pipeline of steganography in real world applications. To address such an issue, we propose a novel scheme for steganography of steganographic networks in this paper. Unlike the existing steganographic schemes which focus on the subtle modification of the cover data to accommodate the secrets. We propose to disguise a steganographic network (termed as the secret DNN model) into a stego DNN model which performs an ordinary machine learning task (termed as the stego task). During the model disguising, we select and tune a subset of filters in the secret DNN model to preserve its function on the secret task, where the remaining filters are reactivated according to a partial optimization strategy to disguise the whole secret DNN model into a stego DNN model. The secret DNN model can be recovered from the stego DNN model when needed. Various experiments have been conducted to demonstrate the advantage of our proposed method for covert communication of steganographic networks as well as general DNN models.

Meiling Li, Nan Zhong, Xinpeng Zhang et al.

Backdoor attack against image classification task has been widely studied and proven to be successful, while there exist little research on the backdoor attack against vision-language models. In this paper, we explore backdoor attack towards image captioning models by poisoning training data. Assuming the attacker has total access to the training dataset, and cannot intervene in model construction or training process. Specifically, a portion of benign training samples is randomly selected to be poisoned. Afterwards, considering that the captions are usually unfolded around objects in an image, we design an object-oriented method to craft poisons, which aims to modify pixel values by a slight range with the modification number proportional to the scale of the current detected object region. After training with the poisoned data, the attacked model behaves normally on benign images, but for poisoned images, the model will generate some sentences irrelevant to the given image. The attack controls the model behavior on specific test images without sacrificing the generation performance on benign test images. Our method proves the weakness of image captioning models to backdoor attack and we hope this work can raise the awareness of defending against backdoor attack in the image captioning field.

Chunxiao Li, Xiaoxiao Wang, Meiling Li et al.

With the rapid advancement of generative models, highly realistic image synthesis has posed new challenges to digital security and media credibility. Although AI-generated image detection methods have partially addressed these concerns, a substantial research gap remains in evaluating their performance under complex real-world conditions. This paper introduces the Real-World Robustness Dataset (RRDataset) for comprehensive evaluation of detection models across three dimensions: 1) Scenario Generalization: RRDataset encompasses high-quality images from seven major scenarios (War and Conflict, Disasters and Accidents, Political and Social Events, Medical and Public Health, Culture and Religion, Labor and Production, and everyday life), addressing existing dataset gaps from a content perspective. 2) Internet Transmission Robustness: examining detector performance on images that have undergone multiple rounds of sharing across various social media platforms. 3) Re-digitization Robustness: assessing model effectiveness on images altered through four distinct re-digitization methods. We benchmarked 17 detectors and 10 vision-language models (VLMs) on RRDataset and conducted a large-scale human study involving 192 participants to investigate human few-shot learning capabilities in detecting AI-generated images. The benchmarking results reveal the limitations of current AI detection methods under real-world conditions and underscore the importance of drawing on human adaptability to develop more robust detection algorithms.